Image-based Phenotyping and Genetic Analysis of Potato ...MATERIALS AND METHODS Plant Material and Field Trials A group of 15 red varieties and advanced breeding lines from the University

Image-based Phenotyping and Genetic Analysis of Potato Skin Set and Color

Maria V. Caraza-Harter, Jeffrey B. Endelman*

Dep. Horticulture, Plant Breeding and Plant Genetics Graduate Program, Univ. Wisconsin-

Madison, Madison, WI 53706. *Corresponding author ([email protected]).

Abbreviations:

BLUP, best linear unbiased predictor; DAP, days after planting; HARS, Hancock Agricultural

Research Station; RGB, red, green and blue; HCL, Hue, Chroma and Lightness.

certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprint (which was notthis version posted July 8, 2019. ; https://doi.org/10.1101/694745doi: bioRxiv preprint

https://doi.org/10.1101/694745

ABSTRACT

Image-based phenotyping offers new opportunities for fast, objective, and reliable measurement

for breeding and genetics research. In the current study, image analysis was used to quantify

potato skin color and skin set, which are critical for the marketability of new varieties. A set of

15 red potato varieties and advanced breeding lines was evaluated over two years at a single

location, with two harvest times in the second year. After mechanical harvest and grading, 7-8

representative tubers per plot were photographed, and the photos were analyzed with ImageJ to

measure skinning (as % surface area) and skin color using the Hue, Chroma and Lightness

(HCL) representation. The plot-based heritability was consistently high (> 0.77) across traits and

environments; the genetic correlation between environments was also high, ranging from 0.81 to

0.98. Significant increases in Lightness and Chroma, as well as a decrease in skinning, were

observed at the late compared to early harvest, while the opposite trends for color were observed

after six weeks of storage. The three color traits were unexpectedly collinear in this study, with

the first principal component explaining 86% of the variation. This result may reflect the

physiology of red color in potato, but the highly selected nature of the 15 genotypes may also be

a factor. Image-based phenotyping offers new opportunities to advance genetic gain and

understanding for tuber appearance traits that have been difficult to precisely measure in the past.


https://doi.org/10.1101/694745

Potato is an important crop for both processing and fresh markets, with the latter category

representing 27% of total US production. The main fresh market types sold in the US are russets,

whites, yellows, and reds. Tuber appearance has a strong impact on marketability and is

therefore important to evaluate during variety development. Historically, potato breeders have

used visual ratings (i.e., 1–5) to score traits that affect tuber appearance, such as length/width

ratio, height/width ratio, curvature, eye depth, skin color, skin finish (i.e., netted vs. smooth), and

skin set (i.e., resistance to excoriation). This approach is labor-intensive, subjective, and often

lacks precision. The alternative of image-based phenotyping of tuber appearance provides an

opportunity to move beyond these limitations. Computer vision has been previously used in

potato and other horticultural crops for grading of produce, such as identifying defects in shape

or color (Patel et al., 2012; Tao et al., 1995). Instead of quality control purposes, our motivation

is to investigate the genetics of skin set and color in red potatoes.

Red skin color in potatoes is due to the presence of anthocyanin pigments in the tuber

periderm, which can be quantified to study the influence of variety and management on color

(Andersen et al., 2002; Hung et al., 1997; Roe et al., 2014; Rosen et al., 2009; Waterer, 2010).

With image-based phenotyping, however, the goal is to measure human perception of color

rather than its chemical basis. A number of mathematical models exist for representing color.

The RGB (red, green, blue) model is widely known and used in digital cameras, but the biconic

Hue, Chroma and Lightness (HCL) model is more closely related to human perception (Figure

1). Hue corresponds to the polar angle, which we have centered on red at 0°, with yellow at 60°

and magenta at -60°. The vertical dimension is Lightness, ranging from 0 (black) to 1 (white).

Chroma is the radial dimension, which ranges from grayscale at 0 to fully saturated at C = 2L.

The HCL color model has been used before in potato based on measurements with a handheld

colorimeter, to study the effects of management, soil type, and storage on a limited number of

varieties (Andersen et al., 2002; Roe et al., 2014; Rosen et al., 2009). The current study builds on

this earlier research by (1) extracting HCL phenotypes from images and (2) examining a larger

set of genotypes.


https://doi.org/10.1101/694745

Figure 1. Biconic geometry of the HCL color model. Image distributed under CC BY-SA 3.0 (Jacob

Rus, SharkD, https://commons.wikimedia.org/wiki/File:HSL_color_solid_dblcone_chroma_gray.png;

changed background to white and text angle).

Potato tuber "skin," or periderm, is composed of three tissues: phellem, phellogen and

phelloderm (Reeve et al., 1969). The phellogen is meristematic (cork cambium) tissue, adding

cells of suberized phellem to the outside and phelloderm tissue to the inside (Lulai et al., 2001).

During the early stages of tuber development, the phellogen layer is active and very susceptible

to excoriation, or "skinning." Skinning not only reduces the marketability of tubers but also the

ability to retain moisture and resist disease during storage. As the tuber matures, changes in the

phellogen layer promote greater adhesion of the phellem—a process informally known as skin

set. According to USDA grading standards, the highest grade of "practically no skinning" means

not more than 5% of the potatoes have more than 10% of the skin missing or feathered (USDA,

2008).

Two different approaches to measuring skin set have been used. One is to measure the

torque at which the periderm excoriates using a torquemeter (Lulai et al., 1993). A more direct

approach, which is amenable to image-based phenotyping and more closely aligned with human

perception, is to record the percent of missing skin on an area basis (Gao et al., 2016). Varietal

differences in skin set are widely recognized, but very little is known about the genetic basis of

this critical trait (Halderson et al., 1993; Lulai, 2007).

The objectives of this study were to (1) use image-based phenotyping to measure skin

color and skin set for a group of 15 commercial varieties and advanced breeding lines; (2)

determine the heritability of the image-based phenotypes, both within and across environments;

and (3) investigate the influence of harvest and storage time on these traits.


https://doi.org/10.1101/694745

MATERIALS AND METHODS

Plant Material and Field Trials

A group of 15 red varieties and advanced breeding lines from the University of

Wisconsin-Madison were evaluated in 2015 and 2016 as replicated 15-plant plots at the UW

Hancock Agricultural Research Station. We evaluated 13 clones in 2015 using a randomized

complete block design (RCBD) with three replications. The experiment was planted April 27 and

harvested 121 Days After Planting. Fertility, water, and pest management followed UW-

Extension guidelines for potato (Bussan et al., 2015). Diquat bromide was applied 14 and 7 days

before harvest to promote vine desiccation. Tubers were mechanically harvested into 30 cm × 45

cm rigid plastic milk crates, run through a washing and grading line, and then crated up again for

storage at 12°C with 95% relative humidity. No additional steps were taken to promote skinning.

In 2016, 11 of the 13 clones from the 2015 trial were evaluated again, plus two additional check

varieties, for a total of 13 clones (Table S1). The 2016 experiment consisted of two adjacent

RCBD trials, each with two replicates, planted on April 21. The first trial was harvested 109

DAP and the second 138 DAP. Crop management and harvest followed the same protocols as

2015.

Image Acquisition and Analysis

Photos were taken within a few days of harvest in 2015 and 2016, as well as six weeks

after harvest in 2016. A set of 7-8 representative tubers from each plot were placed on a black

board and photographed, on one side in 2015 and on both sides in 2016, using a Photosimile 200

Lightbox equipped with a CanonEOS T5i camera (Figure S1). The camera was set to autofocus

with an aperture of F20, an ISO 100 and a shutter speed of 1/10. A Small MacBeth Color Card

was included in each photo in 2016 to compensate for potential variation in lighting and

exposure.

The dataset of 253 images was analyzed using the ImageJ software (Schneider et al.,

2012). For the 2016 photos, the first step consisted of image calibration based on the color card,

using the ImageJ plugin Chart White Balance (Vander Haeghen, 2007). For both years, a semi-

automated background removal was performed using color thresholds to differentiate tubers

from the black background, and the total tuber surface area was measured in pixels. Hue and


https://doi.org/10.1101/694745

brightness thresholds were then used to select and measure the skinned surface area. The ratio

between the skinned area and total tuber surface area is reported as skinning percentage (%) to

quantify skin set. To measure color, hue and brightness thresholds were used to select red skin

and exclude external defects such as exposed tissue (due to skinning) and common scab. The

RGB Measure Plugin was used to measure the average R, G, and B values of the selected area on

a 0–255 scale. RGB values were divided by 255 to fall in the range 0–1 and then converted to the

HCL representation according to the following standard formulas (Smith, 1978):

! = max(', ), *)

, = min(', ), *)

/ = ! −,

1 = ! +,

2

5∗ =

⎩⎪⎪⎨

⎪⎪⎧;?@=,@?/ = 0) − *

/mod6,@?! = '

* − '

/+ 2,@?! = )

' − )

/+ 4,@?! = *

5 = F60° × 5∗,@?5∗ ≤ 3

60° × (5∗ − 6),@?5∗ > 3

Statistical Analysis

Initially, the color and skin set measurements taken at harvest were analyzed separately

for each of the three environments: 2015@121DAP, 2016@109DAP, and 2016@138DAP. The

phenotype Pij for genotype i in block j was modeled by

LMN = O + )M + *N + PMN [1]

where μ is the intercept and )M, *N, and PMN are normally distributed random effects for genotype,

block and residuals, respectively. Variance components were estimated by Restricted Maximum

Likelihood with the ASReml-R software (Butler et al., 2009; R. Core Team, 2018). After

inspecting the residuals, a log transformation was used for skinning % to satisfy the normality

assumption. Plot-based heritability was estimated by


https://doi.org/10.1101/694745

ℎR =STR

STR + SUR

[2]

where STR and SUR are the variance components for genotype and residual, respectively.

For the combined analysis of the at-harvest measurements from the three environments,

we fitted the following mixed model:

LMNV = O + WV + *NV + )WMV + PMNV [3]

In Equation 3, μ is the intercept, WV is the fixed effect of environment, *NVis the random effect of

block nested within environment, GEik is the random effect of genotype i nested within

environment k, and PMNVare residuals. A separable covariance model was used for the GEik effect,

such that the effects for two different genotypes were independent but not the effects for the

same genotype in two different environments:

cov[)WMV, )WM[V[] = ]MM[ΩVV[ [4]

In Equation 4, ]MM[ is the Kronecker delta, which equals 1 when its two arguments are identical

and 0 otherwise, and ΩVV[ is the genetic covariance between environments k and k'. Variance

components and the fixed effects for environment were estimated using ASReml-R. The

statistical significance of pairwise differences between environments was determined with

ASReml-R based on a Wald test and p = 0.05 threshold. The genetic correlation _VV[between

environments k and k' was calculated as

_VV[ =ΩVV[

`ΩVVΩV[V[[5]

)m

Because of the high genetic correlation between environments, a single BLUP (best

unbiased linear predictor) was calculated for each clone using a modification of Equation 3.

The )WMV effect was rewritten as )M + a)WMV to separate the main effect from the G×E interaction,

both of which were assumed to be normally distributed and independent: )M~cd0, SeRf and

)Wa MV~cd0, SRegf. Variance components were estimated with ASReml-R and BLUP[)M] ≡ M was

calculated from the Henderson (1975) mixed model equations. The reliability (nMR) of )mM was

estimated from its prediction error variance (LWoM = opn[)mM − )M]) according to (Clark et al.,

2012)

nMR = 1 − r

LWoMSeRs [6]


https://doi.org/10.1101/694745

Principal component analysis of the three color trait BLUPs, standardized to have zero mean and

unit variance, was performed using the princomp function in R.

The effect of storage time on the color traits was estimated using the 2016 data, based on

the following linear mixed model:

LMNVt = O + )M + *N + WV + ut + )WMV + )uMt + WuVt + )WuMVt + PMNVt [7]

In Equation 7, the intercept is represented by μ; Gi, Bj, and εijk are the random effects for

genotype, block, and residuals respectively. Ek is the fixed effect of environment, with two levels

for the factor (109 and 138 DAP), and Tl is the fixed effect of storage time, with two levels for

the factor (0 and 6 weeks after harvest). All interaction terms were random except ETkl. Because

the effect of storage time was estimated from measurements on the same field plot, a correlated

model for the residuals was used:

vwxyPMNVt, PM[N[V[t[z = ]MM[]NN[]VV[Λtt[ [8]

In Equation 8, ] is the Kronecker delta, and Λ is a 2×2 covariance matrix estimated with

ASReml-R. The statistical significance of the storage time effect (Tl) was determined based on a

Wald test and p = 0.05 threshold. From Equation 7, the intraclass correlation _tt[between the

genotypic values of one clone at different storage times (from the same field environment) is

_tt[ =STR + ST|

R

STR + ST|

R + ST}R + ST|}

R [9]

RESULTS

In the first experiment (in 2015), photographs of 13 red clones were taken within a few

days of harvest (121 DAP) and used to estimate Hue, Chroma, Lightness, and skinning % for

each plot (Figure 2). Hue ranged from -4.6° to 6.5°, Chroma from 0.26 to 0.35, and Lightness

from 0.25 to 0.33. The range for skinning was 2.2% to 30.2%. The plot-based heritability

exceeded 0.75 for all four traits (Table 1). As shown in Figure 2, the three color traits were

highly correlated, but skinning showed only weak or no correlation with color.


https://doi.org/10.1101/694745

Figure 2. Pairwise scatterplot of Hue, Chroma, Lightness and Skinning % for each plot of the 2015@121DAP experiment. Correlation coefficient (r) and p-values shown above the diagonal. Table 1. Plot-based heritability for color and skin set measurements taken at harvest, for three environments (Year@HarvestTime) in Wisconsin.

Trait Environment 2015@121DAP 2016@109DAP 2016@138DAP

Hue 0.82 0.82 0.82 Chroma 0.91 0.95 0.87 Lightness 0.91 0.96 0.95 Skinning 0.83 0.77 0.86


https://doi.org/10.1101/694745

The experiment was repeated for a second season, in 2016, with separate trials for early

(109 DAP) and late (138 DAP) harvests. The plot-based heritability for both harvest times was

similar to the 2015 experiment (Table 1). From a combined analysis of the three environments

(2015@121DAP, 2016@109DAP, 2016@138DAP), the statistical significance of the

environment effect was estimated (Table 2). Hue was significantly higher and Lightness

significantly lower in 2015 compared to 2016. Looking at the effect of harvest time in 2016,

there was no significant difference in Hue, while Chroma and Lightness were both higher for the

late harvest. Skinning was not significantly different between 2015 and the early 2016 harvest,

but less skinning was observed in the late 2016 harvest.

Table 2. Environment means for the color traits and skin set. Means with different letters are significantly different based on a Wald test with p < 0.05.

Trait Environment 2015@121DAP 2016@109DAP 2016@138DAP

Hue (°) 1.02a -2.57b -2.35b Chroma 0.32a 0.32a 0.34b Lightness 0.29a 0.33b 0.35c Skinning (%) 5.89a 4.99a 3.26b

Despite the significant main effect of environment, there was very little G×E in this

experiment. The genetic correlation between environments exceeded 0.8 for all four traits and all

three pairwise comparisons (Table 3). This allowed for the calculation of a single BLUP per

clone across the three environments (Table S2). For the clones evaluated in both years, the

reliability of the BLUPs exceeded 0.7 for all four traits (Table S2). Because of the high

correlation between the three color traits (Figure 1), a principal component (PC) analysis of the

BLUPs was performed. The first PC captured 83% of the variation (Figure S2), with loadings of

0.57 for Hue and Chroma and 0.60 for Lightness.


https://doi.org/10.1101/694745

Table 3. Genetic correlations for color and skin set between three Wisconsin environments, based on measurements taken at harvest.

Trait 2015@121DAP–2016@109DAP 2015@121DAP–2016@138DAP

2016@109DAP–2016@138DAP

Hue 0.81 0.83 0.98 Chroma 0.87 0.82 0.88 Lightness 0.85 0.93 0.96 Skinning 0.89 0.96 0.84

The first PC for color was plotted against skinning % to visualize the genetic variation for

this set of 15 clones (Figure 3). The top two red varieties in Wisconsin, as well as the entire US,

are Red Norland and Dark Red Norland (National Potato Council, 2018), which are line

selections (i.e., somatic mutants) of Norland (Johansen et al., 1959). A major reason for the

continued dominance of Norland selections is resistance to skinning, which is consistent with

their position along the horizontal axis in Figure 3. As the name suggests, Dark Red Norland was

darker than Red Norland in our experiment, but several breeding lines (e.g., W8893-1R, W6511-

1R) were even darker. The potential for large differences even among close relatives is

illustrated by the full-sibs W10209-2R and W10209-7R, for which the skinning percentages

were 3.2% and 17.1%, respectively.

Figure 3. Scatterplot of BLUPs for color (PC1) vs. skinning % for the 15 red clones evaluated in three environments.


https://doi.org/10.1101/694745

The effect of storage time on color was estimated using the 2016 data, for which

photographs were taken at 0 and 6 weeks after harvest. For all three color traits, the plot-based

heritability remained high after six weeks of storage (≥ 0.88). There was also very little G×E

between the two storage times, with genetic correlations above 0.9 for all three traits (Table S3).

The color traits were significantly affected by storage time: Hue increased by 3o, Chroma

decreased by 0.03, and Lightness decreased by 0.03 (Table 4). The perceived effect of harvest

and storage time is visible in Figure 4, which compares images of the variety 'Red Prairie' at

different harvest times and before and after storage. The BLUPs for each clone at each harvest

and storage time are provided in Table S4.

Figure 4. The effects of harvest time (109 vs 138 DAP) and storage time (0 vs. 6 weeks) for the variety 'Red Prairie.' Storage led to decreased Chroma and Lightness, while the opposite trends were observed for the late vs. early harvest. Besides the effect of storage on skin color, the images show changes in the color of the skinned area and higher severity of skin blemish diseases (e.g., silver scurf and black dot).


https://doi.org/10.1101/694745

Table 4. The effect of storage time on color, from the 2016 experiment. Means with different letters are significantly different based on a Wald test with p < 0.05.

Trait At Harvest 6 Weeks Hue (°) -2.49a 0.77b Chroma 0.33a 0.30b Lightness 0.34a 0.31b

DISCUSSION

The primary motivation for this research was to develop an image-based phenotyping

method for tuber appearance that can be used for breeding and genetics research. Image-based

phenotyping is ubiquitous now due to the availability of multi-spectral sensors on UAVs (Li et

al., 2019), but imaging studies of plant morphology are also becoming more common (Darrigues

et al., 2008; Miller et al., 2017; Moore et al., 2013). Plot-based heritability (h2) is a critical

measure of the reliability of the phenotyping method, and we were pleased to estimate values

over 0.75 for all three color traits and skinning percentage. Because h2 was similar between the

early and late harvest in 2016, and because there was very little G×E between these

environments (genetic correlations exceeded 0.8), it appears the timing of harvest (within reason)

is not critical for selection or genetic mapping for these traits.

Potato growers often refer to the "loss of color" that occurs during storage, which negatively

impacts the marketability of red potatoes. In this experiment, "loss of color" manifested as lower

Chroma and lower Lightness at 6 weeks after harvest compared to right after harvest (see Table

4). Figure 4 illustrates these changes in tuber appearance for one genotype. Previous studies on

the effect of storage on red color, based on measurements with a handheld colorimeter, have also

reported decreases in Chroma and/or Lightness (Andersen et al., 2002; Roe et al., 2014; Rosen et

al., 2009). Selecting genotypes that maintain Chroma in storage is an important breeding goal,

but there was very little G×E for this trait (genetic correlations exceeded 0.9). Since the 15

clones in this experiment are representative of the genetic diversity of the UW-Madison red

breeding program, new germplasm may be needed to make genetic gains for color retention.

Compared with storage time, studies on the effect of harvest time on red skin color are rarer

and less consistent. In this study, Lightness and Chroma were significantly higher for tubers

harvested 138 DAP compared to 109 DAP. Rosen et al. (2009) measured skin color at harvest


https://doi.org/10.1101/694745

compared to vine kill in two years, reporting decreased Lightness and no change in Chroma in

the first year but higher Lightness and lower Chroma in the second year.

The physiological basis for the changes in tuber appearance reported here deserves further

study. Hung et al. (1997) reported that both Chroma and anthocyanidin (the aglycone form of

anthocyanin) content per unit surface area decreased during tuber growth (i.e., "bulking"); the

authors hypothesized this was due to pigment dilution (from increased surface area) and/or

degradation. Sulc et al. (2017) measured anthocyanidin content in potatoes with pigmented skin

and flesh (which are a specialty item, not a major commodity, in the US), reporting a fairly

consistent decline over a 15-week period. Extrapolating these results to our study, we would

predict there to be less anthocyanin in the late-harvest tubers compared to early-harvest, and yet

the late-harvest tubers had higher Chroma. Both Andersen et al. (2002) and Roe et al. (2014)

reported decreases in anthocyanin content during storage, which seems consistent with our

finding of lower Chroma.

The strong collinearity between the three color traits in this study was unexpected. The HCL

color model is three-dimensional, but for the 15 genotypes in this study, the color variation was

largely one-dimensional (the first PC explained 86% of the variation). This result may reflect the

biology of red color in potato, but the highly selected nature of the 15 genotypes in this study

may be a factor. Support for the latter hypothesis comes from an ongoing genetic mapping

project in which hundreds of unselected F1 progeny from the UW-Madison red potato breeding

program have been imaged, and for which the color traits are less correlated (data not shown).

The genetics of red skin color as a qualitative (presence/absence) trait is well characterized (Jung

et al., 2009; Zhang et al., 2009), but our understanding of color as a quantitative trait, particularly

in tetraploid potato, is incomplete. Much less is known about the genetics of skin set, as there

have been only a few studies based on gene expression (Neubauer et al., 2013; Vulavala et al.,

2017) and none based on association or linkage analysis.

ACKNOWLEDGMENTS

Financial support was provided by the Wisconsin Department of Agriculture Specialty Crop

Block Grant (16-02), the Wisconsin Potato and Vegetable Growers Association, and the UW

Office of the Vice Chancellor for Research and Graduate Education. L. Snodgrass, G.

Christensen, and B. Kleven assisted with the harvest and photographing of tubers.


https://doi.org/10.1101/694745

REFERENCES

Andersen, A.W., C. Tong and D.E. Krueger. 2002. Comparison of periderm color and anthocyanins of four red potato varieties. American Journal of Potato Research 79: 249-253.

Bussan, A.J., V.M. Colquhoun, A.J. Davis, R.L. Gevens, D.J. Groves, B.M. Heider, et al. 2015. Commercial Vegetable Porduction in Wisconsin.

Butler, D., B. Cullis, A. Gilmour and B. Gogel. 2009. ASReml-R reference manual Version 3, Queensland Department of Primary Industries and Fisheries, Brisbane.

Clark, S.A., J.M. Hickey, H.D. Daetwyler and J.H.J. Van der Werf. 2012. The importance of information on relatives for the prediction of genomic breeding values and the implications for the makeup of reference data sets in livestock breeding schemes. Genet Sel Evol 44: 1-9.

Darrigues, A., J. Hall, E. Van Der Knaap and D.M. Francis. 2008. Tomato Analyzer-color Test: A New Tool for Efficient Digital Phenotyping. Journal of American Society of Horticulture 133: 579-586.

Gao, Y., J. Geng, X. Rao and Y. Ying. 2016. CCD-Based skinning injury recognition on potato tubers (Solanum tuberosum L.): A comparison between visible and biospeckle imaging. Sensors 16.

Halderson, J.L. and R.C. Henning. 1993. Measurements for determining potato tuber maturity. American Potato Journal 70: 131-141.

Henderson, C.R. 1975. Best Linear Unbiased Estimation and Prediction under a Selection Model. Biometrics 31: 423-447.

Hung, C., J. Murray, S. Ohmann and C. Tong. 1997. Anthocyanin accumulation during potato tuber development. Journal of American Society of Horticulture 122: 20-23.

Johansen, R.H., N. Sandar, W.G. Hoyman and E.P. Lana. 1959. Norland a new red-skinned potato variety with early maturity and moderate resistance to common scab. American Potato Journal 36: 12-15.

Jung, C., H. Griffiths, D. De Jong, S. Cheng, M. Bodis, T. Kim, et al. 2009. The potato developer (D) locus encodes an R2R3 MYB transcription factor that regulates expression of multiple anthocyanin structural genes in tuber skin. Theoretical and Applied Genetics 120: 45-57. doi:10.1007/s00122-009-1158-3.

Li, B., X. Xu, J. Han, L. Zhang, C. Bian, L. Jin, et al. 2019. The estimation of crop emergence in potatoes by UAV RGB imagery. Plant Methods 15: 15.

Lulai, E.C. 2007. The canon of potato science: 43. Skin-set and wound-healing/suberization. Potato Research 50: 387-390.

Lulai, E.C. and T.P. Freeman. 2001. The Importance of Phellogen Cells and their Structural Characteristics in Susceptibility and Resistance to Excoriation in Immature and Mature Potato Tuber (Solanum tuberosum L.) Periderm. Analysis of Botany 88: 555-561.

Lulai, E.C. and P.H. Orr. 1993. Determining the Feasibility of Measuring Genotyping Differences in Skin-Set. American Journal of Potato Research 70: 599-609.

Miller, N.D., N. Haase, J. Lee, S. Kaeppler, N. De Leon and E. Spalding. 2017. A robust, high-throughput method for computing maize ear,cob, and kernel attributes automatically from images. The Plant Journal 89: 169-178.

Moore, C., L. Johnson, I.-Y. Kwak, M. Livny, K. Broman and E. Spalding. 2013. High-Throughput Computer Vision Introduces the Time Axis to a Quantitative Trait Map of a Plant Growth Response. Genetics 195: 1077-1086.

National Potato Council. 2018. Potato Statistical Yearbook 2018.


https://doi.org/10.1101/694745

Neubauer, J., E. Lulai, A. Thompson, J. Suttle, M. Bolton and L. Campbell. 2013. Molecular and cytological aspects of native periderm maturation in potato tubers. Journal of Plant Physiology 170: 413-423.

Patel, K.K., A. Kar, S.N. Jha and M.A. Khan. 2012. Machine vision system: a tool for quality inspection of food and agricultural products. Journal of Food and Science Technology 49: 123-141.

R. Core Team. 2018. R: A language and environment for statistical computing. Viena, Austria. Reeve, R.M., E. Hautala and M.L. Weaver. 1969. Anatomy and compositional variation whitin

potatoes. American Journal of Potato Research 46: 361-373. Roe, M., J. Carlson, T. McManimon, A. Hegeman and C. Tong. 2014. Differential Accumulation

and Degradation Of Anthocyanins In Red Norland Periderm is Dependent On Soil Type And Tuber Storage Duration. American Journal of Potato Research 91: 696-705. doi:10.1007/s12230-014-9402-z.

Rosen, C., J.A. Roessler, P.D. Petracek, C.B.S. Engelman and C. Tong. 2009. 2,4-Dichlorophenoxyacetic acid increases peonidin derivates in Red Norland periderm. American Journal of Potato Research 86: 15-23.

Schneider, C., W. Rasband and K. Eliceiri. 2012. NIH Image to ImageJ: 25 years of image analysis. Nature Methods 9.

Smith, A. 1978. Color gamut transform pairs. Computer Graphics 12: 12-19. Sulc, M., Z. Kotikova, L. Paznocht, V. Pivec, K. Hamouz and J. Lachman. 2017. Changes in

anthocyanidin levels during the maturation of color-fleshed potato (Solanum tuberosum L.) tubers. Food Chemistry 237: 981-988.

Tao, Y., P. Heinemann, Z. Varghese, C. Morrow and H. Soomer III. 1995. Machine vision for color inspection of potatoes and apples. American Society of Agricultural Engineers 38: 1555-1561.

USDA, U.S.D.o.A. 2008. United States Standards for Grades of Potatoes. In: A. M. Service, editor.

Vander Haeghen, Y. 2007. Chart_White_Balance, C4Real. Realistic, web-ready sRGB pictures from your digital camera.

Vulavala, V., E. Fogelman, L. Rozental, A. Faigenboim, Z. Tanami, O. Shoseyoc, et al. 2017. Indetification of genes related to skin development in potato. Plant Molecular Biology 94: 481-494.

Waterer, D. 2010. Influence of growth regulators on skin colour and scab diseases of red-skinned potatoes. Canadian Journal of Plant Science 90: 745-753.

Zhang, Y., C.S. Jung and W.S. De Jong. 2009. Genetic analysis of pigmented tuber flesh in potato. Theoretical and Applied Genetics 119: 143.


https://doi.org/10.1101/694745

SUPPLEMENTAL MATERIAL

Image-based Phenotyping and Genetic Analysis of Potato Skin Set and Color

Caraza-Harter and Endelman

Figure S1. Imaging system at the Hancock Agricultural Research Station in Wisconsin.


https://doi.org/10.1101/694745

Figure S2. Biplot from a principal component analysis of the BLUPs for the color traits, across all three environments.


https://doi.org/10.1101/694745

Table S1. List of red potato varieties and advanced breeding lines evaluated in 2015 and 2016 at the Hancock Agricultural Research Station in Wisconsin. Clone Environment

2015 2016 121 DAP 109 DAP 138 DAP

Chieftain NA * * DarkRedNorland * * * NDW102738CB-1R * NA NA RedEndeavor * * * RedLaSoda10 * * * RedNorland NA * * RedPrairie * * * VillettaRose * * * W10114-3R * * * W10209-2R * * * W10209-7R * * * W6511-1R * * * W8886-3R * NA NA W8890-1R * * * W8893-1R * * *

* : Evaluated NA: Not Available


https://doi.org/10.1101/694745

Table S2. BLUPs and reliabilities across all three environments for the two-year experiment. Clone Hue (°) Chroma Lightness Skinning (%)

BLUP r2 BLUP r2 BLUP r2 BLUP r2 Chieftain 3.35 0.77 0.33 0.84 0.35 0.88 7.15 0.82 DarkRedNorland -2.16 0.82 0.34 0.87 0.33 0.90 2.55 0.86 NDW102738CB-1R 0.97 0.68 0.35 0.77 0.32 0.84 15.31 0.76 RedEndeavor -1.01 0.82 0.32 0.87 0.33 0.90 5.94 0.86 RedLaSoda10 0.88 0.82 0.34 0.87 0.34 0.90 5.37 0.86 RedNorland 0.38 0.78 0.34 0.84 0.38 0.88 2.31 0.83 RedPrairie 0.39 0.82 0.35 0.87 0.36 0.90 4.47 0.86 VillettaRose -2.67 0.82 0.31 0.87 0.32 0.90 3.99 0.86 W10114-3R -1.14 0.82 0.35 0.87 0.33 0.90 2.48 0.86 W10209-2R -3.46 0.82 0.29 0.87 0.29 0.90 3.20 0.86 W10209-7R -0.41 0.82 0.32 0.87 0.32 0.90 17.05 0.86 W6511-1R -4.00 0.82 0.29 0.87 0.28 0.90 3.07 0.86 W8886-3R -3.01 0.68 0.32 0.77 0.31 0.84 4.19 0.76 W8890-1R -3.22 0.82 0.31 0.87 0.29 0.90 2.22 0.86 W8893-1R -4.70 0.82 0.32 0.87 0.29 0.90 3.24 0.86

Table S3. Variance components for clone (G), harvest environment (E, 109 vs. 138 DAP), and storage time (T, 0 vs. 6 weeks) in the 2016 experiment. Trait Variance components

VG VGT VGE VGET

Hue 5.50 0.00 0.91 0.00 Chroma 5.50 0.00 0.91 0.00 Lightness 14.95 0.92 0.70 0.00


https://doi.org/10.1101/694745

Table S4. Genotype BLUPs for each combination of harvest and storage time in the 2016 experiment. Clone Hue (°) Chroma Lightness Storage time At Harvest 6 weeks At Harvest 6 weeks At Harvest 6 weeks

109 DAP

138 DAP

109 DAP

138 DAP

109 DAP

138 DAP

109 DAP

138 DAP

109 DAP

138 DAP

109 DAP

138 DAP

Chieftain 1.64 3.39 5.09 5.96 0.33 0.35 0.31 0.33 0.36 0.38 0.32 0.36 DarkRedNorland -5.14 -5.38 -1.92 -2.47 0.34 0.36 0.31 0.33 0.34 0.36 0.30 0.34 RedEndeavor -1.61 -1.80 2.18 1.53 0.32 0.34 0.30 0.31 0.33 0.36 0.30 0.35 RedLaSoda10 0.62 0.59 4.26 4.15 0.33 0.35 0.32 0.34 0.34 0.38 0.31 0.35 RedNorland -1.16 -0.11 2.59 2.56 0.35 0.35 0.33 0.33 0.39 0.40 0.34 0.37 RedPrairie -1.43 -0.74 1.82 2.53 0.35 0.37 0.33 0.34 0.37 0.40 0.32 0.37 VillettaRose -4.19 -3.05 -0.63 0.12 0.31 0.32 0.29 0.29 0.32 0.34 0.30 0.33 W10114-3R -3.70 -2.59 -0.13 0.09 0.35 0.37 0.31 0.33 0.33 0.36 0.30 0.34 W10209-2R -3.82 -4.61 -0.35 -1.56 0.29 0.31 0.26 0.28 0.29 0.31 0.27 0.30 W10209-7R -2.04 -0.55 1.46 2.05 0.30 0.33 0.28 0.30 0.30 0.33 0.27 0.31 W6511-1R -3.20 -4.81 0.78 -2.03 0.27 0.29 0.24 0.26 0.28 0.30 0.26 0.28 W8890-1R -4.32 -4.61 -1.04 -1.24 0.30 0.33 0.28 0.30 0.29 0.31 0.27 0.30 W8893-1R -5.32 -6.70 -1.82 -3.92 0.31 0.34 0.27 0.31 0.29 0.31 0.27 0.30


https://doi.org/10.1101/694745

Documents

Image-based Phenotyping and Genetic Analysis of Potato ...MATERIALS AND METHODS Plant Material and Field Trials A group of 15 red varieties and advanced breeding lines from the University